Joseph Lim

A Systems Design Engineering student at the University of Waterloo with a keen interest in Data Science and Machine Learning

ARC-AGI Challenge 2024

This project involved experimenting with LLMs to address the ARC-AGI Challenge, introduced by François Chollet in 2019. It focused on developing specialized models by fine-tuning LLaMA language models on augmented clusters of similar puzzle tasks using Unsloth, resulting in notable performance improvements during early evaluations

Auto Differentiation Package

This project involved developing an autodifferentiation package from scratch using Numpy, supporting multi-layer fully-connected networks with different nonlinearities for both regression and binary classification tasks.

K-pop Song Recommender

This project was designed to delve into the intricacies of K-pop music by extracting song features from a range of K-pop artists to construct a Content-Based Recommendation System. Utilizing the Spotify API, a robust data pipeline was developed to gather the necessary information, forming a comprehensive dataset. The recommendation engine was then crafted, applying principles of linear algebra to generate and present top song suggestions to users based on their song preferences.

NBA Game Winner Predictor

This project aimed to predict the outcomes of NBA games by analyzing the past 7 seasons and applying various machine learning models on the data. The project involved all stages of the machine learning process, including data collection, preprocessing, and model development. With a focus on optimizing predictive accuracy, the models achieved a test accuracy of 70%.